1081: Running Back Prospects Who Produced Like Studs

**JJ Zachariason** (0:02)
This is The Late-Round Podcast with your host, JJ Zachariason.
What's up, everyone? It's JJ Zachariason. In this episode, 1081 of The Late-Round Fantasy Football Podcast, thanks for tuning in. Before getting to the topic today, I just want to send another reminder that The Late-Round Prospect Guide is currently available for preorder. This year's guide lets you know everything that goes into the ZAP models. It's got profiles for every running back, wide receiver and tight end going to the NFL combine. And then there's also the year two profiles. I have my year two model for second year NFL players. And I've got a little over 30 profiles for running back some wide receivers there, too. So this thing is a beast, and you can preorder it for $14.99. That's going to go up to $19.99 when the guide launches in mid-March. Just head to lateround.com to learn more. Today's an episode that I do each year as sort of an unofficial kickoff to my prospecting work. For those who are unfamiliar, I have a prospect model or models, depending on how you want to frame it. That's what I just talked about at the top. I've worked on them for years, and I try to improve them every year with new data and new ways of calculating that data. I used to be pretty bad at modeling. Like the first couple of iterations weren't that great in hindsight, but I've worked really hard over the last seven or eight years to get these models to a place where I'm confident in these results. I know they bring value because I've tested them pretty thoroughly. With any model, you should be aiming to forecast something. In the case of the ZAP models, at least at Running Back and Wide Receiver, I'm aiming to predict a player's best two season average and PPR points per game across their first three years in the league. I call that B2S or best two seasons. Think of it this way. A player plays 15 games in each of his first three seasons. So he's hit a minimum game threshold. In year one, he scores 11 PPR points per game. In year two, it's 12 In year three, that jumps up to 14 That player's B2S would be 13 PPR points per game. Because it's taking his year three number of 14 and it's averaging that with his second highest output, which is 12 PPR points per game. That rookie season average just gets thrown out. I'm attempting to predict B2S for every running back and wide receiver who either goes to the combine or gets drafted. And I've got data that goes back to 2011 to help me do that. So the ZAP model database is filled with hundreds of running backs and wide receivers. Some players do slip through the cracks, like Austin Eckler, who wasn't at the combine or drafted. Adam Thielen, same exact deal. The models have a variety of inputs, ranging from a player's size to teammate score to production-related metrics. The production side is what we're going to be focused on today. Now, some people think collegiate production just doesn't matter when you're evaluating incoming prospects. And it might not matter at certain positions. I'm not analyzing linebackers and safeties here. But at running back and wide receiver, collegiate production matters, and it can matter a lot. Part of the reason people think production in college doesn't matter is because they just think in terms of raw production. Yards, receptions, touchdowns. Those things can be really skewed in college given the variety of offenses and schemes that we see. One team might throw the ball 200 times in the season. Another team might throw it 700 times. That's not what these models really look at, though. The production metrics in the ZAP model, they're share-related. That is, they're related to how a player performed within his own offense. We look at those kind of metrics all the time. Target share, touchdown share. The list goes on and on. On top of that, we can take those share-focused metrics and adjust them for things like age and program strength. A player giving us crazy production within his offense as a freshman is more likely going to be better than a player giving us that same production as a fifth-year senior. And if he does that at Alabama, that's more impressive than if he did it at Alabama State.
Now, clearly, these models are being tested against what we've seen historically. And since these production metrics are predictive, that means players in the past at these positions share some similar traits. Turns out, productive NFL players were almost always productive in college. Not all productive college players are then productive in the NFL. But if a player is dominating as a pro, he was probably dominating in college too. That brings me back to today's show. Every year is sort of a primer. I look at a sample of historical running backs and wide receivers who have been studs in the NFL. A stud sample, if you will. I find what model-related traits those guys have in common. And then I see if there are any players from this year's class who produced like them in college. I see if anyone in this year's class produced like an NFL stud. Today, we're talking running backs. Next week, we're going to go to wide receivers. Okay, let's get to the good stuff now. There's really no simple way to determine what an NFL running back stud is. Is it someone who gives us 16 PPR points per game in any season of his career? Is it someone who gives us multiple RB2 seasons across some portion of his career? What is it? My definition just for purposes of the show and to make some sense of this class is any running back since 2011 who had more than one season with 14 or more PPR points per game. That should obviously capture the studs like Christian McCaffrey, but it gives us a larger sample of players who may have been rock-solid RB2s for multiple years too. Not everyone in that sample is in the ZAP model database, so I can't compare their metrics to this year's class. But when those guys are all filtered out, it still gives me 39 running backs, 39 stud NFL running backs to compare this year's class to. So everyone's aware there are 20 running backs that were invited to this year's combine. Technically, it's 21, but one's a very obvious and clear fullback, so we're removing him from the sample. The goal now is to compare their production profiles to the NFL stud sample and their production profiles, their collegiate production profiles. The running back model changed a bit year over year, but as usual, it's very receiving heavy. I've said this before, but that's likely due to a few things. Number one, receiving might just be a little bit undervalued in general. Draft capital is an input in the model, and that could be doing a lot of the heavy lifting for the rushing numbers. Number two, the model is building for PPR leagues, so receptions are going to matter a little bit more. And then number three, receiving is a talent signal. When a player has a wider skill set, that tells us he might be pretty talented. The three main production metrics that currently go into the running back zap model, their pro rated best season reception share, adjusted yards per team play, and adjusted receiving yards per team pass attempt. Reception share is simply running back receptions divided by total team receptions. The percentage of receptions that a running back had for his team. That's pro rated for games played, and a player must have played at least six games in the season for that shared account. Next is adjusted yards per team play. This takes total yards by a running back. It divides it by the number of plays his team ran. The adjusted piece is a multiplier. It's adjusting for age and strength of schedule. So, the younger a player, the harder the schedule, the larger that multiplier is going to be for total yards per team play. That exact same logic is used for receiving yards per team pass attempt. Except with receiving yards per team pass attempt, you're looking at receiving yards per team pass attempt. This is a metric that I used last year in the model for running backs, but I did it in the form of breakout score. I'm now only using breakout score for wide receivers. More on that next week though. Now, at this point in the show, someone's out there saying, get on with it, JJ. Just tell me who produced like a stud. Because look, it happens every single year. But folks, welcome to The Late-Round Podcast. Process is the focus here. Anyway, let's look at that stud sample. When it came to pro-rated reception share in their best collegiate season, the stud sample averaged a 13.7% share. That actually was very similar to this year's class. But unfortunately, that's because of Eli Heidenreich. He's a running back out of Navy. Navy historically has a run-heavy offense, and that was no different this past year. They had just 106 completions all year, and Heidenreich caught 51 of those balls. He had a 48% reception share. That's absolutely breaking the model. Now, I've read that some people think that he might be a fullback, but who knows how a team is going to deploy that kind of player at the NFL level. But without his crazy reception share, the average in this class drops to 11.8%. And for the record, he's not the fullback that I was talking about earlier. But an 11.8% reception share for this year's class, that's not that bad. And just to be clear, I'm going to remove Heidenreich from the averages moving forward, because he's really skewing this data. Now, when it comes to adjusted yards per team play, the stud sample is 2.12. This year's class, one that's not filled with studs, 1.60. Makes sense. And then lastly, with adjusted receiving yards per team pass attempt, it's 0.98 for the stud sample. With this year's combine invites, 0.70.